Monday, July 20: The EU forces Google to open Android's AI layer to rivals; Gemini 3.5 Pro misses its third deadline as Google eyes a Flash stopgap; NVIDIA ships Cosmos 3 Edge and pulls 22 Japanese robotics giants into a physical-AI coalition; and new data shows Anthropic and OpenAI now claim 89 percent of AI startup revenues. In this hour: EU Orders Google to Open Android's AI Layer — Apple Pulls Siri AI From Europe Rather Than Comply · Gemini 3.5 Pro Misses Third Deadline — Google Confirms It's Testing a Stopgap 'Upgraded Flash' · NVIDIA Launches Cosmos 3 Edge On-Device World Model — 22 Japanese Giants Join Physical-AI Coalition · Anthropic and OpenAI Now Claim 89% of AI Startup Revenues — Anthropic's Run-Rate Tops OpenAI's.

Monday, July 20: The EU's DMA forces Google to open Android to rival AI assistants as Apple quietly pulls new Siri AI from Europe; Anthropic's Fable 5 enters Max and Team Premium subscriptions today; Google's Gemini 3.5 Pro misses a third launch deadline after a coding-data training reset proved disappointing; and Kimi K3's open weights are locked in for July 27. In this hour: Brussels Forces Open Android's AI Layer — Rivals Get the Keys Gemini Kept · Fable 5 Goes Live in Max and Team Premium — With a Catch on Limits · Gemini 3.5 Pro Misses Third Deadline — Coding Reset Backfired, Stopgap Eyed · Kimi K3 Open Weights Drop July 27 — Frontier-Class Model Goes Fully Open.

Monday, July 20: Kimi K3's open weights land July 27 as US labs reassess China's coding lead; Google confirms a Flash-tier stopgap in partner testing while Gemini 3.5 Pro slips further; Microsoft's Patch Tuesday fixes a record 570 vulnerabilities with AI help as CISA flags autonomous agents as a new identity threat; and a buried FTC filing reveals SpaceX quietly acquired a 1-gigawatt gas-turbine firm to power xAI's Colossus 2. In this hour: Kimi K3 Confirms Free Weights July 27 — US Labs Reassess China's Coding Lead · Google Confirms Upgraded Flash Stopgap in Partner Testing as Gemini 3.5 Pro Keeps Slipping · Patch Tuesday Fixes Record 570 Vulnerabilities With AI Help — CISA Warns Agents Are the Next Identity Gap · Buried FTC Filing Reveals SpaceX Bought a 1-Gigawatt Gas-Turbine Firm for $1B to Power xAI's Colossus 2.

Monday, July 20: Moonshot's Kimi K3 blows past GPU limits — subscriptions paused, IPO revealed; Alibaba drops Qwen3.8-Max-Preview the same weekend; Gemini 3.5 Pro slips again dragging Alphabet shares; OpenAI confirms GPT-5.6 Sol deletes user files under certain conditions; San Francisco forces Apple and Google to pull AI nudify apps. In this hour: Kimi K3 Maxes Out Moonshot's GPUs — Subscriptions Paused, IPO Revealed · Alibaba Drops Qwen3.8-Max-Preview — 2.4 Trillion Parameters, Same Weekend as Kimi · Gemini 3.5 Pro Misses Again — Alphabet Shares Slide ~4% · OpenAI Confirms GPT-5.6 Sol Deletes User Files — Misconfigured Environment Variables to Blame · San Francisco Orders Apple and Google to Pull AI Nudify Apps — Both Comply.

Monday, July 20: Kimi K3 hits frontier-tier rankings as open weights approach; Google misses Gemini 3.5 Pro's third launch window; DeepSeek's V4 API cutover deadline tightens on developers; Japan locks in a ¥387B sovereign AI commitment; and the UK AI Security Institute puts a number on how fast China is closing the open-weight gap. In this hour: Kimi K3 Lands at #4 Globally — Open Weights Due July 27 · Gemini 3.5 Pro Misses July 17 — Coding Fix Disappoints, Flash Stopgap in Testing · DeepSeek V4 Migration Mandatory by Thursday — Legacy Aliases Go Dark July 24 · Japan Commits ¥387B to Sovereign AI — Targets Operations by June 2028.

Monday, July 20: Google's Gemini 3.5 Pro is months behind schedule on coding; Japan launches a $2.4B physical-AI factory on Nvidia Rubin chips; and OpenAI teases a cooperative-subagent tier of GPT-5.6 Sol Ultra for Codex. In this hour: Gemini 3.5 Pro Is Months Late — Coding Shortfall, Internal Strain, and a Missed June Deadline · OpenAI Teases GPT-5.6 Sol Ultra for Codex — Cooperative Subagents That Talk Mid-Task · Japan Launches $2.4B Physical-AI Factory on Nvidia Rubin — 44 Companies, 10 Million Robots by 2040.

Monday, July 20: Moonshot raises at a $31.5B valuation as Kimi K3 open weights near release; Japan commits ¥387.3B to sovereign AI compute; DeepSeek's legacy API aliases hit a hard July 24 cutoff; Huawei's Atlas 950 SuperPoD draws fresh independent scrutiny; and Reddit reports AI spam moderation now catching 25,000 posts a day. In this hour: Moonshot Eyes $31.5B Valuation as Kimi K3 Open Weights Near Release · Japan Locks In ¥387.3B for AI Datacenters — Operations Targeted for June 2028 · Huawei's Atlas 950 SuperPoD Claims 6.7× Nvidia's NVL144 — Independent Scrutiny Builds · DeepSeek's Legacy API Aliases Die Thursday — Four Days to Migrate · Reddit Catches 25,000 AI-Spam Posts Per Day — GEO Targeting ChatGPT Citations Is the New Vector.

Monday, July 20: SK Group's Chey warns AI memory has become a matter of national economic security; China's CXMT files for an $8.6B IPO to challenge the memory big three; Samsung unveils HBM4; Trump moves to weaponize the UN against EU AI rules; and South Korea's FSC clamps down on leveraged chip ETFs as SK Hynix shares reel. In this hour: SK Group's Chey: Governments Are Weaponizing AI Memory Access · China's CXMT Files $8.6B IPO — The Memory Big Three Has a Fourth Contender · Samsung Unveils HBM4 — Staking Its Claim on the Next AI Memory Generation · Seoul Triples Margins on Chip ETFs as SK Hynix Shares Reel 40% From Peak · Trump to Push UN 'Free Expression' Declaration Aimed at EU's AI Rules.

Monday, July 20: WAIC 2026 closes in Shanghai with WAICO's 29-nation governance body crystallising; Japan and Nvidia lock in a 140MW Rubin AI factory for physical AI; Hugging Face reveals its breach forensics hit a wall when US frontier-model safety guardrails blocked its own tools. In this hour: WAIC 2026 Closes With WAICO Constituted — 29 Nations, No Western Democracies, and a $147B Industry Figure · Japan and Nvidia Stand Up a 140MW Rubin AI Factory — 27,500 GPUs, ¥1 Trillion, Open Weights Required · Hugging Face Used Open-Weight GLM-5.2 for Breach Forensics — US Frontier Model Safety Guardrails Blocked the Investigation · Two Tracks, No Bridge — WAICO and FRONTia Signal a World Where Nations Build Before They Agree.

Monday, July 20: Illinois becomes the first US state to mandate annual third-party safety audits of frontier AI companies; OpenAI teases a compute-intensive Sol Ultra tier for Codex; Meta's Watermelon model reportedly matches GPT-5.5 on benchmarks; Feyn Labs ships a text-to-SQL model family; global VC hits a record $510B in H1 2026 with AI dominating; and NAVER locks in a gigawatt-scale sovereign AI factory with Nvidia in South Korea. In this hour: Illinois Signs First US Law Mandating Annual Safety Audits of Frontier AI Labs · OpenAI Teases GPT-5.6 Sol Ultra — A Compute-Intensive Tier Above the Current Flagship · Scale AI's Wang Says Meta's 'Watermelon' Has Matched GPT-5.5 — Still in Training · Global VC Hits Record $510B in H1 2026 — OpenAI and Anthropic Alone Take $217B · NAVER and Nvidia Lock In a Gigawatt-Scale Sovereign AI Factory in South Korea · Feyn Labs Ships SQRL — A Text-to-SQL Model That Probes Before It Queries.